[daip] Imagr problem

[Eric Greisen]

I have gotten your data and have gotten IMAGR to go wrong - on a
Solaris machine where the debugger is better.  I will pursue this
going wrong business: at around 10000 components, after a major cycle,
it decides to take an enormous number from one of the outer fields
about 9000 CCs from 6000 pixels loaded to the AP.  Curiously, that is
about the number of pixels in the window.  Then it switches to field
1, where it loads > 900000 pixels to the AP, again close to the window
size.  This grossly ignores maxpixel and causes the Clark algorithm to
run ordres of magnitude more slowly.  I suspect that it would have
eventually returned from this cycle - it was adding to the CC file
slowly - but I killed it.  So something is really going wrong here -
controls and the like appear to be overwritten.  This takes hours to
get to the problem.  Sigh... I guess we now have a higher class of
bugs.

I do have some questions: do you really mean for all the outer fields
to be 2048x2048 with a tiny cleaning window (only 25x256)?  I used the
TV and noticed that the main central sources appear in every image.
In that case, you really have to use at least OVERLAP=1.  Otherwise,
they are subtracted out of every field but restored only to the field
(1) in which they were found.  OVERLAP=2 would be better still on this
image since you want Clean to ignore the outer fields for quite a
while since I saw no real sources in the (small) Clean windows
although there MAY be real sources outside the small windows in the
outer fields.  The blow-up of Clean may have to do with the state of
its residual histogram - the images looked like nothing but noise
before it went into its bad state.

Thanks for following this up - I have had Clean go nuts once before
and we could not figure out what happened.  This may help us do so.

Eric